Vector Optimized Library of Kernels 2.5.1
Architecture-tuned implementations of math kernels
 
Loading...
Searching...
No Matches
volk_16i_max_star_16i.h
Go to the documentation of this file.
1/* -*- c++ -*- */
2/*
3 * Copyright 2012, 2014 Free Software Foundation, Inc.
4 *
5 * This file is part of GNU Radio
6 *
7 * GNU Radio is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 3, or (at your option)
10 * any later version.
11 *
12 * GNU Radio is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with GNU Radio; see the file COPYING. If not, write to
19 * the Free Software Foundation, Inc., 51 Franklin Street,
20 * Boston, MA 02110-1301, USA.
21 */
22
53#ifndef INCLUDED_volk_16i_max_star_16i_a_H
54#define INCLUDED_volk_16i_max_star_16i_a_H
55
56#include <inttypes.h>
57#include <stdio.h>
58
59#ifdef LV_HAVE_SSSE3
60
61#include <emmintrin.h>
62#include <tmmintrin.h>
63#include <xmmintrin.h>
64
65static inline void
66volk_16i_max_star_16i_a_ssse3(short* target, short* src0, unsigned int num_points)
67{
68 const unsigned int num_bytes = num_points * 2;
69
70 short candidate = src0[0];
71 short cands[8];
72 __m128i xmm0, xmm1, xmm3, xmm4, xmm5, xmm6;
73
74 __m128i* p_src0;
75
76 p_src0 = (__m128i*)src0;
77
78 int bound = num_bytes >> 4;
79 int leftovers = (num_bytes >> 1) & 7;
80
81 int i = 0;
82
83 xmm1 = _mm_setzero_si128();
84 xmm0 = _mm_setzero_si128();
85 //_mm_insert_epi16(xmm0, candidate, 0);
86
87 xmm0 = _mm_shuffle_epi8(xmm0, xmm1);
88
89 for (i = 0; i < bound; ++i) {
90 xmm1 = _mm_load_si128(p_src0);
91 p_src0 += 1;
92 // xmm2 = _mm_sub_epi16(xmm1, xmm0);
93
94 xmm3 = _mm_cmpgt_epi16(xmm0, xmm1);
95 xmm4 = _mm_cmpeq_epi16(xmm0, xmm1);
96 xmm5 = _mm_cmpgt_epi16(xmm1, xmm0);
97
98 xmm6 = _mm_xor_si128(xmm4, xmm5);
99
100 xmm3 = _mm_and_si128(xmm3, xmm0);
101 xmm4 = _mm_and_si128(xmm6, xmm1);
102
103 xmm0 = _mm_add_epi16(xmm3, xmm4);
104 }
105
106 _mm_store_si128((__m128i*)cands, xmm0);
107
108 for (i = 0; i < 8; ++i) {
109 candidate = ((short)(candidate - cands[i]) > 0) ? candidate : cands[i];
110 }
111
112 for (i = 0; i < leftovers; ++i) {
113 candidate = ((short)(candidate - src0[(bound << 3) + i]) > 0)
114 ? candidate
115 : src0[(bound << 3) + i];
116 }
117
118 target[0] = candidate;
119}
120
121#endif /*LV_HAVE_SSSE3*/
122
123#ifdef LV_HAVE_NEON
124#include <arm_neon.h>
125
126static inline void
127volk_16i_max_star_16i_neon(short* target, short* src0, unsigned int num_points)
128{
129 const unsigned int eighth_points = num_points / 8;
130 unsigned number;
131 int16x8_t input_vec;
132 int16x8_t diff, zeros;
133 uint16x8_t comp1, comp2;
134 zeros = vdupq_n_s16(0);
135
136 int16x8x2_t tmpvec;
137
138 int16x8_t candidate_vec = vld1q_dup_s16(src0);
139 short candidate;
140 ++src0;
141
142 for (number = 0; number < eighth_points; ++number) {
143 input_vec = vld1q_s16(src0);
144 __VOLK_PREFETCH(src0 + 16);
145 diff = vsubq_s16(candidate_vec, input_vec);
146 comp1 = vcgeq_s16(diff, zeros);
147 comp2 = vcltq_s16(diff, zeros);
148
149 tmpvec.val[0] = vandq_s16(candidate_vec, (int16x8_t)comp1);
150 tmpvec.val[1] = vandq_s16(input_vec, (int16x8_t)comp2);
151
152 candidate_vec = vaddq_s16(tmpvec.val[0], tmpvec.val[1]);
153 src0 += 8;
154 }
155 vst1q_s16(&candidate, candidate_vec);
156
157 for (number = 0; number < num_points % 8; number++) {
158 candidate = ((int16_t)(candidate - src0[number]) > 0) ? candidate : src0[number];
159 }
160 target[0] = candidate;
161}
162#endif /*LV_HAVE_NEON*/
163
164#ifdef LV_HAVE_GENERIC
165
166static inline void
167volk_16i_max_star_16i_generic(short* target, short* src0, unsigned int num_points)
168{
169 const unsigned int num_bytes = num_points * 2;
170
171 int i = 0;
172
173 int bound = num_bytes >> 1;
174
175 short candidate = src0[0];
176 for (i = 1; i < bound; ++i) {
177 candidate = ((short)(candidate - src0[i]) > 0) ? candidate : src0[i];
178 }
179 target[0] = candidate;
180}
181
182#endif /*LV_HAVE_GENERIC*/
183
184
185#endif /*INCLUDED_volk_16i_max_star_16i_a_H*/
static void volk_16i_max_star_16i_neon(short *target, short *src0, unsigned int num_points)
Definition: volk_16i_max_star_16i.h:127
static void volk_16i_max_star_16i_a_ssse3(short *target, short *src0, unsigned int num_points)
Definition: volk_16i_max_star_16i.h:66
static void volk_16i_max_star_16i_generic(short *target, short *src0, unsigned int num_points)
Definition: volk_16i_max_star_16i.h:167
#define __VOLK_PREFETCH(addr)
Definition: volk_common.h:62
for i
Definition: volk_config_fixed.tmpl.h:25